Misinterpretations of results in weed research through inappropriate use of statistics

Summary Examples of weed research experiments are described where inappropriate statistical analyses have commonly been used and, as a result, unjustifiable conclusions were reached. The examples are for herbicide performance trials, crop yield/weed density studies, time of onset of weed competition and ‘critical period’ studies. Erroneous conclusions were drawn because non‐transformed data were used and multiple comparisons were made with methods such as Duncan's Multiple Range Test and the use of a Least Significant Difference. More thorough analyses, especially using graphs and regressions, are shown to lead to more reliable interpretations, which are quite different from those usually concluded. In particular, considerable treatment effects are revealed in cases where the original conclusions were that they were absent.